# GPT 5.1 Thinking

The deliberation-focused variant in the GPT-5.1 family, applying extended chain-of-thought reasoning to complex problems where thoroughness and accuracy outweigh response speed.

- **Model ID:** `openai/gpt-5.1-thinking`
- **Type:** chat
- **Providers:** openai, azure
- **Context window:** 400,000
- **Maximum output tokens:** 128,000
- **Pricing:** $1.25/1M input tokens, $10/1M output tokens
- **Canonical page:** https://vercel.com/ai-gateway/models/gpt-5.1-thinking

## Supported parameters

Detailed capability metadata has not been reported for this model.

## Example

```ts
import { streamText } from 'ai'

const result = streamText({
  model: 'openai/gpt-5.1-thinking',
  prompt: 'Why is the sky blue?'
})
```

## About

GPT 5.1 Thinking was released on November 12, 2025 as part of the GPT-5.1 model generation on AI Gateway. It occupies the reasoning-depth end of the GPT-5.1 spectrum, complementing the speed-optimized instant variant.

The model uses extended chain-of-thought computation, generating internal reasoning tokens that work through problems step by step before producing a visible response. This approach, proven effective in the o-series models, is here applied within the GPT-5.1 architecture to produce more accurate and thorough answers on complex problems.

The context window of 400K tokens supports the lengthy inputs that complex reasoning tasks often require, whether they are research papers, detailed problem statements, or extensive codebases that need architectural analysis. Longer response times come in exchange for deeper, more reliable reasoning.

## What to consider

GPT 5.1 Thinking generates internal reasoning tokens before producing its final response, similar to the o-series reasoning models but within the GPT-5.1 architecture. This produces better results on hard problems at the cost of longer response times.

Use GPT 5.1 Thinking for your hardest queries and GPT-5.1 instant for everything else. A routing layer can direct traffic based on query complexity.

## When to use

### Best For

- **Complex analysis:** Multi-step research, data analysis, and strategic reasoning that benefits from deliberation
- **Mathematical problem solving:** Proofs, derivations, and quantitative analysis requiring verified steps
- **Scientific reasoning:** Physics, chemistry, and biology problems with multi-step logical chains
- **Hard coding problems:** Algorithm design, optimization, and architectural decisions requiring extended thought
- **High-stakes decision support:** Analysis where accuracy matters more than speed

### Consider Alternatives When

- **Fast responses needed:** GPT-5.1 instant for real-time interactions
- **Coding agent workflows:** GPT-5.1 codex family for autonomous software engineering
- **Pure STEM reasoning:** The o-series reasoning models for the deepest chain-of-thought capability in OpenAI's lineup
- **General chat:** GPT-5 chat or GPT-5.1 instant for conversational workloads where reasoning depth is unnecessary

## Best for

- **Complex analysis:** Multi-step research, data analysis, and strategic reasoning that benefits from deliberation
- **Mathematical problem solving:** Proofs, derivations, and quantitative analysis requiring verified steps
- **Scientific reasoning:** Physics, chemistry, and biology problems with multi-step logical chains
- **Hard coding problems:** Algorithm design, optimization, and architectural decisions requiring extended thought
- **Extended chain-of-thought:** The GPT-5.1 generation's deliberation tier, complementing the speed-optimized instant variant

## Consider alternatives

- **Fast responses needed:** GPT-5.1 instant for real-time interactions
- **Coding agent workflows:** GPT-5.1 codex family for autonomous software engineering
- **Pure STEM reasoning:** The o-series reasoning models for the deepest chain-of-thought capability in OpenAI's lineup
- **General chat:** GPT-5.1 instant for conversational workloads where reasoning depth is unnecessary

## Frequently asked questions

### How does GPT 5.1 Thinking reasoning work?

It generates internal reasoning tokens that work through the problem step by step before producing a visible response, similar to the approach used in o-series reasoning models.

### When should I use thinking versus instant?

Use thinking for complex analysis, math, science, and hard coding problems where accuracy is the priority. Use instant for real-time chat, streaming content, and tasks where speed is the priority.

### Is GPT 5.1 Thinking slower than GPT-5.1 instant?

Yes. The extended reasoning process adds time before the first visible output. The tradeoff is deeper, more accurate reasoning on complex problems.

### What context window does GPT 5.1 Thinking support?

400K tokens, supporting the lengthy inputs that complex reasoning tasks often require.

### How does AI Gateway handle authentication for GPT 5.1 Thinking?

AI Gateway accepts a single API key or OIDC token for all requests. You don't embed OpenAI credentials in your application; AI Gateway routes and authenticates on your behalf.

### What are typical latency characteristics?

This page shows live throughput and time-to-first-token metrics measured across real AI Gateway traffic.

## Links

- [Model page](https://vercel.com/ai-gateway/models/gpt-5.1-thinking)
- [AI Gateway documentation](https://vercel.com/docs/ai-gateway)
- [Provider model documentation](https://openai.com/index/gpt-5-1/)
- [Provider pricing](https://platform.openai.com/docs/pricing)